CN101568014A

CN101568014A - Estimating device and method, and program

Info

Publication number: CN101568014A
Application number: CNA2009101336907A
Authority: CN
Inventors: 小林由幸
Original assignee: Sony Corp
Current assignee: Sony Corp
Priority date: 2008-04-22
Filing date: 2009-04-22
Publication date: 2009-10-28
Anticipated expiration: 2029-04-22
Also published as: US8160987B2; JP2009265729A; JP4591793B2; CN101568014B; US20090265294A1

Abstract

An estimating device includes: a predictive computing unit configured to estimate, based on an evaluation matrix made up of an evaluation value indicating an evaluation as to each of multiple evaluation targets for each of multiple users, and an estimated expression for estimating the evaluation value by computation employing the evaluation matrix, the evaluation value of the evaluation target which has not been subjected to an evaluation by the user, and obtain a predictive evaluation value which is the estimated evaluation value; and a linear combining unit configured to subject a plurality of the predictive evaluations obtained by employing a plurality of the estimated expressions to linear combination by employing a linear combination coefficient, thereby obtaining a final estimation result of an evaluation as to the evaluation target which has not been subjected to the evaluation by the user.

Description

Assessment apparatus and method and program

Technical field

The present invention relates to a kind of assessment apparatus and method and program, and particularly, relate to the assessment apparatus and method and the program that are fit to be applied under the situation of user's content recommendation.

Background technology

Up to now, according to prior art, exist and assess the technology of user for the evaluation of predetermined content by collaborative filtering.

Particularly, for the messaging device of using this collaborative filtering, there is a kind of messaging device, this messaging device is checked by application and is listened to similar other users' of interested user historical content and carry out the collaborative filtering processing, with the evaluation (for example, see Japanese unexamined patent open No.2005-167628) of assessment for each content of interested user.By this messaging device, in a plurality of contents, the content with evaluation of the highest assessment is used as interested user institute Cup of tea thing and recommends this user.

Summary of the invention

Yet,, be difficult to accurately assess the evaluation of user for content by above-mentioned technology.

For example, for the messaging device of applicating cooperation filter algorithm, just use and check and listen to the similar user's of interested user history and assess evaluation for interested user's content.Therefore, the assessment about for the evaluation of content does not wherein reflect the generally evaluation for this content, i.e. all evaluations of users, and therefore, the evaluation that is obtained by assessment may greatly differ from each other with interested user's authentic assessment.

Have realized that and to make that the user can be by accurate assessment for the evaluation of content.

According to embodiments of the invention, a kind of assessment apparatus comprises: prediction and calculation unit, be configured to assess not the evaluation of estimate of carrying out the evaluation objective estimated by the user based on the evaluation matrix that constitutes for each the evaluation of estimate of evaluation in a plurality of evaluation objectives by expression each among a plurality of users and assessment expression formula, and obtain the prediction and evaluation value as the evaluation of estimate of assessment, the assessment expression formula is used for assessing evaluation of estimate by the calculating of applicating evaluating matrix; And linear combination unit, be configured to by using linear combination coefficient, obtain thus for the final assessment result of not carrying out the evaluation of the evaluation objective estimated by the user to carrying out linear combination by using a plurality of prediction and evaluation that a plurality of assessment expression formulas are obtained.

The assessment expression formula can be made of a plurality of operators, and described operator comprises carries out the operator that collaborative filtering calculates.

Assessment apparatus can also comprise: assessment expression formula candidate generation unit, be configured to a plurality of assessment expression formulas as assessment expression formula candidate, assessment expression formula candidate is the candidate of more recent application in a plurality of assessment expression formulas of calculating final assessment result, and this assessment expression formula candidate generation unit generates new arbitrarily assessment expression formula and assesses the new assessment expression formula of the part acquisition of expression formulas and be used as assessing the expression formula candidate by revising in a plurality of assessment expression formulas some; The assessment result generation unit, be configured to, at each assessment expression formula candidate, based on assessment expression formula candidate and evaluation matrix, calculate to estimate each user in the matrix for each the prediction and evaluation value in a plurality of evaluation objectives, generate the evaluation result that constitutes by the prediction and evaluation value that obtains by calculating; The linear combination coefficient computing unit, be configured to some the assessment results in a plurality of assessment results as use, and obtain linear combination coefficient by using the assessment result of using and estimating matrix the assessment expression formula candidate who use to use under as the situation of assessment expression formula, the assessment expression formula candidate of use is the assessment expression formula candidate who is used to generate the evaluation result of use; Evaluation unit is configured to calculate the amount of information benchmark of conduct for the evaluation of assessment expression formula candidate who uses and linear combination coefficient; And selected cell, the assessment expression formula candidate of use who is configured to have in the assessment expression formula candidate that select to use according to the amount of information benchmark and the linear combination coefficient high praise is as newly being applied to a plurality of assessment expression formulas and the linear combination coefficient that final assessment result is calculated.

The linear combination coefficient computing unit in a plurality of users that belong to a group in a plurality of groups, can be used and belongs to and the user's of described group of identical group evaluation of estimate and the linear combination coefficient that the prediction and evaluation value obtains each group.

According to embodiments of the invention, a kind of appraisal procedure or program may further comprise the steps: assess not the evaluation of estimate of being carried out the evaluation objective estimated by the user based on the evaluation matrix that is made of for each the evaluation of estimate of evaluation in a plurality of evaluation objectives expression each among a plurality of users and assessment expression formula, and obtain the prediction and evaluation value as the evaluation of estimate of assessment, the assessment expression formula is used for assessing evaluation of estimate by the calculating of applicating evaluating matrix; And, obtain thus for the final assessment result of not carrying out the evaluation of the evaluation objective estimated by the user by using linear combination coefficient to carrying out linear combination by using a plurality of prediction and evaluation values that a plurality of assessment expression formulas obtain.

Pass through the foregoing description, estimate not the evaluation of estimate of carrying out the evaluation objective estimated by the user based on the evaluation matrix that constitutes for each the evaluation of estimate of evaluation in a plurality of evaluation objectives by expression each among a plurality of users and assessment expression formula, the assessment expression formula is used for assessing evaluation of estimate by the calculating of applicating evaluating matrix, and obtain prediction and evaluation value as the evaluation of estimate of assessment, and, obtain thus for the final assessment result of not carrying out the evaluation of the evaluation objective estimated by the user by using linear combination coefficient to carrying out linear combination by using a plurality of prediction and evaluation values that a plurality of assessment expression formulas obtain.

According to embodiments of the invention, the user can be assessed more accurately for the evaluation of content.

Description of drawings

Fig. 1 is the diagram that the overview of the content recommendation system of using embodiments of the invention is described;

Fig. 2 is the diagram that the overview of the content recommendation system of using embodiments of the invention is described;

Fig. 3 shows the block diagram of the ios dhcp sample configuration IOS DHCP of content recommendation system;

Fig. 4 shows the block diagram of the more detailed ios dhcp sample configuration IOS DHCP of prediction and evaluation value computing unit;

Fig. 5 shows the block diagram of the more detailed ios dhcp sample configuration IOS DHCP of unit;

Fig. 6 is the flow chart that the explanation user estimates the renewal processing of matrix;

Fig. 7 shows the diagram of the example of assessment expression formula;

Fig. 8 shows the diagram of the example of operator;

Fig. 9 A to 9C shows the diagram of the example of assessment expression formula;

Figure 10 is the flow chart of explanation recommendation process;

Figure 11 is the diagram of concrete example of the calculating of explanation operator;

Figure 12 is the diagram of concrete example of the calculating of explanation operator;

Figure 13 is the diagram of concrete example of the calculating of explanation operator;

Figure 14 is the diagram of concrete example of the calculating of explanation operator;

Figure 15 is the diagram of concrete example of the calculating of explanation operator;

Figure 16 is the diagram of concrete example of the calculating of explanation operator;

Figure 17 is the diagram of the generation of the interim prediction matrix of explanation;

Figure 18 is the diagram of the generation of explanation user prediction matrix;

Figure 19 is the flow chart that explanation study is handled;

Figure 20 is the diagram of explanation to user's grouping;

Figure 21 is the flow chart that explanation assessment candidate generates processing;

Figure 22 A to 22F is the diagram of the example of explanation sudden change processing;

Figure 23 is the diagram of the example of explanation cross processing;

Figure 24 is the flow chart of explanation evaluation process;

Figure 25 is the diagram of explanation assessment expression formula candidate's calculating;

Figure 26 is the diagram of the generation of explanation assessment result;

Figure 27 is the flow chart that explanation assessment expression formula is selected processing; And

Figure 28 shows the block diagram of the ios dhcp sample configuration IOS DHCP of computer.

Embodiment

Using embodiments of the invention will describe with reference to accompanying drawing following.The overview of the content recommendation system of having used embodiments of the invention at first, is described with reference to Fig. 1 and Fig. 2.

For this content recommendation system, for example shown in Figure 1, registered in advance a plurality of users and content.Notice that in Fig. 1, the C1 to CM on the horizontal direction represents content, and the U1 to UN on the vertical direction represents the user.That is to say that for content recommendation system, M content C1 to CM and N user U1 to UN are by registered in advance.Each user will represent the evaluation of estimate of the evaluation of content is input to some registered contents.For example, we say that the evaluation of content is divided into five stages, promptly detest, dislike, it doesn't matter, like and like best, and represent that the numerical value 1-5 of each evaluation is used as the evaluation of estimate input.Here, we say that evaluation of estimate is big more, and the user is just high more for the evaluation of content.

For example, in Fig. 1, user U4 estimates for content C1 and C5, and input " 5 " is as the evaluation of estimate of content C5.That is to say, can find user U4 thinking very highly for content C5.Noting, can be the target of any kind as the content of user's evaluation objective, and for example moving image or books etc. are as long as this target is based on the evaluation objective that user's hobby is estimated.

Like this, after by the evaluation of estimate of each user's input, obtain expression for the information of each user to the evaluation of estimate of content for content.Each user of thus obtained expression is maintained at content recommendation system for the information of the evaluation of estimate of content and estimates matrix as the user.

When the acquisition user estimates matrix, the evaluation of estimate that content recommendation system is imported before estimating in the matrix based on the user predicts that (assessment) also do not have the evaluation of estimate of the content of input, and generates by the prediction and evaluation matrix by predicting that prediction and evaluation value that obtains and the evaluation of estimate of importing before constitute.

Here, the evaluation of estimate of input is meant the evaluation of estimate of being imported by the user before.For example, in Fig. 1, " 1 " is used as user U1 and imports for the evaluation of estimate of content C1, and this evaluation of estimate " 1 " as before the evaluation of estimate of input.Equally, for following explanation,, promptly will not be called as the not evaluation of estimate of input for the value of blank column among Fig. 1 by the evaluation of estimate of user's input.For example, therefore user U2 does not exist for the in fact input of the evaluation of estimate of content C1, but user U2 will be called as the not evaluation of estimate of input for the evaluation of estimate that is not transfused to of content C1.

Like this, content recommendation system prediction user estimates the not evaluation of estimate of input in the matrix, and after obtaining the prediction and evaluation value that predicts the outcome as it, obtains the prediction and evaluation matrix that is made of prediction and evaluation value and the evaluation of estimate imported before thus.Therefore, for example, the prediction and evaluation matrix shown in Fig. 2 is estimated the matrix from the user shown in Fig. 1 and is obtained.Notice that in Fig. 2, the evaluation of estimate in the shaded bar is represented the prediction and evaluation value.For example, for the prediction and evaluation matrix among Fig. 2, be set to the prediction and evaluation value " 2 " that obtains by prediction for the evaluation of estimate of the content C1 of user U2.

When the prediction and evaluation value of value that obtains to estimate each blank column of matrix about the user, promptly not the input evaluation of estimate, and when obtaining the prediction and evaluation matrix, the information that content recommendation system provides expression to be assessed as the favorite content of user to predetermined user is to recommend this content.

For example, under the situation of user U2 content recommendation, content recommendation system is given user U2 with the commending contents that has the maximum predicted evaluation of estimate in the registered content.That is to say that in content C1 to CM, the prediction and evaluation value of user U2 is that the content C5 of maximum " 5 " is assessed as the favorite content of user, and the recommended user U2 that gives.

Therefore, for example shown in Fig. 3, configuration is used for estimating matrix from the user and obtains the prediction and evaluation matrix commending contents is given user's commending system.In Fig. 3, content recommendation system is configured to interface treatment facility 11 and recommendation apparatus 12, and interface treatment facility 11 and recommendation apparatus 12 interconnect.

Interface treatment facility 11 is by user's operation, and execution is handled about user's interface.For example, interface treatment facility 11 will be provided to recommendation apparatus 12 by the evaluation of estimate of user input, show that perhaps the expression that provides from recommendation apparatus 12 recommends the information of user's content.

Recommendation apparatus 12 generates the user based on the evaluation of estimate that provides from interface treatment facility 11 and estimates matrix, and estimates the matrix generation forecast based on the user and estimate matrix, and gives the user with commending contents.Recommendation apparatus 12 comprises that unit 21 is accepted in input, the user estimates matrix holding unit 22, unit 23, assessment expression formula holding unit 24, prediction and evaluation value computing unit 25, prediction and evaluation matrix holding unit 26 and recommendation unit 27.

The input that user's evaluation of estimate is accepted in unit 21 is accepted in input.That is to say that input is accepted unit 21 and will be provided to the user from the evaluation of estimate that interface treatment facility 11 provides and estimate matrix holding unit 22.The user estimates matrix holding unit 22 and estimates matrix and keep this matrix based on accepting the evaluation of estimate that unit 21 provides from input and generating the user.More specifically, at every turn when input is accepted unit 21 evaluation of estimate is provided, the user estimates matrix holding unit 22 and the evaluation of estimate that provides is provided is upgraded maintained user and estimate matrix.Equally, the user estimates matrix holding unit 22 and maintained user is estimated matrix is provided to unit 23 and prediction and evaluation value computing unit 25.

Unit 23 user who estimates matrix holding unit 22 and provide from the user is provided is estimated matrix and generate assessment expression formula and the linear combination coefficient of using when the matrix generation forecast is estimated matrix estimating from the user, and these assessment expression formulas and linear combination coefficient are provided to assessment expression formula holding unit 24.For unit 23, generate one or more assessment expression formulas and at each user's linear combination coefficient.Assess assessment expression formula and linear combination coefficient that 24 maintenances of expression formula holding unit provide from unit 23, and maintained assessment expression formula and linear combination coefficient are provided to prediction and evaluation value computing unit 25.

25 application of prediction and evaluation value computing unit are estimated matrix and are calculated the evaluation of estimate of not importing in the matrix of prediction and evaluation value estimate to(for) the user from the assessment expression formula of assessing expression formula holding unit 24 and linear combination coefficient from the user that the user estimates matrix holding unit 22, and this prediction and evaluation value is provided to prediction and evaluation matrix holding unit 26.

Prediction and evaluation matrix holding unit 26 is based on from the prediction and evaluation value of prediction and evaluation value computing unit 25 and generation forecast is estimated matrix, and keeps this prediction and evaluation matrix, and maintained prediction and evaluation matrix is provided to recommendation unit 27.Recommendation unit 27 is searched for the content that will recommend predetermined user based on the prediction and evaluation matrix that provides from prediction and evaluation matrix holding unit 26, and will represent to be provided to interface treatment facility 11 by the recommendation information of the content of Search Results acquisition.

Fig. 4 shows the block diagram of the detailed ios dhcp sample configuration IOS DHCP of the prediction and evaluation value computing unit 25 among Fig. 3.Prediction and evaluation value computing unit 25 comprises interim predicting unit 51 and predicting unit 52, and the user estimates matrix and the assessment expression formula is estimated matrix holding unit 22 from the user and assessment expression formula holding unit 24 is provided to interim predicting unit 51, and linear combination coefficient is provided to predicting unit 52 from assessment expression formula holding unit 24.

Here, the assessment expression formula is to be used for (for example, passing through collaborative filtering) to carry out in order to obtaining the not operation expression of the calculating of the prediction and evaluation value of the evaluation of estimate of input.In other words, be used for the operation expression that predetermined appraisal procedure by the applicating cooperation filter algorithm obtains the prediction and evaluation value and be used as the assessment expression formula.Equally, the linear combination coefficient at each user corresponding to the assessment expression formula is at the coefficient of using when carrying out linear combination according to predicting the outcome of a plurality of assessment expression formulas.

The user that interim predicting unit 51 application provide estimates matrix and the assessment expression formula is calculated the interim prediction and evaluation value that the user estimates the evaluation of estimate of not importing in the matrix, and should interim prediction and evaluation value be provided to predicting unit 52.More specifically, interim predicting unit 51 will be estimated the identical matrix of matrix (information) as interim prediction matrix with the user, and application assessment expression formula is carried out the calculating for interim prediction matrix.Subsequently, interim predicting unit 51 information that will be made of the evaluation of estimate for each content of a user of interim prediction matrix is as interim user's prediction matrix.

Therefore, interim user's prediction matrix is made of evaluation of estimate and the interim prediction and evaluation value for input before a user's the content, and interim prediction matrix is made of interim user's prediction matrix of each user.

For example, if we say that the user who provides among Fig. 1 estimates matrix, then interim predicting unit 51 promptly for content C3 and content C5-CM, is used the interim prediction and evaluation value that an assessment expression formula Fk obtains user U1 for the not evaluation of estimate of input of user U1.Subsequently, interim predicting unit 51 information that will be made of the interim prediction and evaluation value of the evaluation of estimate of input before content C1, C2 and the C4 and content C3 and C5-CM is provided to predicting unit 52 as the interim user's prediction matrix for the user U1 of assessment expression formula Fk.

Therefore, for example, K assessment expression formula F1 to Fk is being provided under the situation of interim predicting unit 51, about (interim user's prediction matrix Tnk of 1≤k≤K) (1≤k≤K wherein, the 1≤n≤N) be provided to predicting unit 52 wherein of each obtained each the assessment expression formula Fk among the user U1 to UN from interim predicting unit 51.That is to say that N * K interim user's prediction matrix Tnk is provided to predicting unit 52 altogether.

Predicting unit 52 application obtain final prediction and evaluation value from the interim prediction and evaluation value of interim predicting unit 51 and the linear combination coefficient that provides from assessment expression formula holding unit 24, and this final prediction and evaluation value is provided to prediction and evaluation matrix holding unit 26.More specifically, interim user's prediction matrix to each user carries out linear combination to predicting unit 52 by each user is used linear combination coefficient, and will be provided to prediction and evaluation matrix holding unit 26 as user's prediction matrix of thus obtained result.

For example, (wherein 1≤n≤N) the individual interim user's prediction matrix Tn1 to Tnk of the K of acquisition carries out linear combination by the linear combination coefficient of user Un, and user's prediction matrix An of user Un is provided thus about user Un.Subsequently, N altogether the user's prediction matrix A1 to AN that obtains about N user is provided to prediction and evaluation matrix holding unit 26, and generates the prediction and evaluation matrix that is made of these user's prediction matrixs.

Secondly, Fig. 5 shows the block diagram of the more detailed ios dhcp sample configuration IOS DHCP of the unit 23 among Fig. 3.Unit 23 generates by genetic programming and is used for the assessment expression formula and the linear combination coefficient that are more suitable for that generation forecast is estimated matrix.In other words, unit 23 makes assessment expression formula and linear combination coefficient optimization by genetic programming applicating cooperation filter algorithm, set up proposed algorithm thus, be used for to be assessed as the favorite commending contents of user and give this user in laborsaving mode.

Unit 23 comprises candidate's generation unit 91, assessment result computing unit 92, group generation unit 93 and assessment expression formula selected cell 94, and estimates matrix from the user that the user estimates matrix holding unit 22 and be provided to assessment result computing unit 92 by assessment expression formula selected cell 94.

Candidate's generation unit 91 K assessment expression formula F1 to Fk providing from assessment expression formula selected cell 94 is provided is assessed expression formula candidate J1 to JS (k≤S) wherein to generate, this J1 to JS is new assessment expression formula candidate, and the assessment expression formula candidate that these are new is provided to assessment result computing unit 92.That is to say that candidate's generation unit 91 is used S the assessment expression formula candidate of the assessment expression formula generation of genetic programming the past generation as follow-on assessment expression formula candidate.Candidate's generation unit 91 comprises sudden change processing unit 111, cross processing unit 112 and random process unit 113.

The assessment expression formula that sudden change processing unit 111 provides from assessment expression formula selected cell 94 by application is carried out sudden change and is handled, and generates thus and assesses the expression formula candidate.Here, sudden change is handled and is meant the processing that generates a new assessment expression formula by a part of revising an assessment expression formula.

112 application of cross processing unit are carried out cross processing from the assessment expression formula that assessment expression formula selected cell 94 provides, and generate assessment expression formula candidate thus.Here, cross processing means the processing that generates a new assessment expression formula by a part of the part of an assessment expression formula being changed into another assessment expression formula.That is to say that for cross processing, the part of an assessment statement formula is replaced by the part of another assessment expression formula.

Random process is carried out in random process unit 113, generates assessment expression formula candidate thus.Here, random process means the processing that the operator of selecting by combining random generates a new assessment expression formula.That is to say,, under the situation of the assessment expression formula of not using last generation, generate new assessment expression formula candidate arbitrarily according to random process.

Candidate's generation unit 91 will be from the assessment expression formula in the last generation that assessment expression formula selected cell 94 provides, and S altogether of the new assessment expression formula that is generated by sudden change processing, cross processing and random process offers assessment result computing unit 92 as assessment expression formula candidates.

Assessment result computing unit 92 is used that the user who estimates matrix holding unit 22 and provide from the user estimates matrix and is come from the assessment expression formula candidate that candidate's generation unit 91 provides that (wherein 1≤s≤S) carries out the assessment that the user estimates the evaluation of estimate of matrix, and (wherein 1≤s≤S) is provided to assessment expression formula selected cell 94 with its assessment result Hs for each assessment expression formula candidate Js.

That is to say, assessment result computing unit 92 will be estimated the identical matrix of matrix (information) as (matrix) assessment result for an assessment expression formula candidate with the user, and use this assessment expression formula candidate and assess evaluation of estimate for each content of each user in the assessment result (user estimates matrix).Subsequently, assessment result computing unit 92 will obtain the assessment result of each evaluation of estimate as the final assessment result for applied assessment expression formula candidate by assessment.Therefore, quantity is provided to assessment expression formula selected cell 94 for the assessment result of assessment expression formula candidate's quantity from assessment result computing unit 92.

Group generation unit 93 user who estimates matrix holding unit 22 and provide from the user is provided is estimated matrix and carry out grouping about the user U1 to UN of registered in advance, and the group information that will represent each user's group and belong to the user of this user's group is provided to and assesses expression formula selected cell 94.

For example, user's brief introduction is provided to group generation unit 93 in appropriate circumstances, and described user's brief introduction is the information at age, sex, address, favorite type of each user of expression or the like.Group generation unit 93 the user's brief introduction that is provided suitably is provided is carried out grouping, makes among the user U1 to UN each all belong to the user that predetermined user organizes among the G1 to GQ and organizes.

94 application of assessment expression formula selected cell are estimated matrix, are generated follow-on assessment expression formula and linear combination coefficient from the assessment result Hs of assessment result computing unit 92 and from user's group information of organizing generation unit 93 from the user that the user estimates matrix holding unit 22, and these follow-on assessment expression formulas and linear combination coefficient are provided to candidate's generation unit 91 and assessment expression formula holding unit 24.

Assessment expression formula selected cell 94 comprises combination coefficient computing unit 114, evaluation unit 115 and selected cell 116.

Combination coefficient computing unit 114 hypothesis is selected some in S the assessment result, and each combination that is used for obtaining the assessment expression formula candidate of each selected assessment result is employed as assessing expression formula.Subsequently, combination assessment expression formula candidate is used as under the situation of assessment expression formula, and combination coefficient computing unit 114 user who is provided is provided is estimated linear combination coefficient that matrix, assessment result and user's group information obtains each user's group as being linear combination coefficient candidate's linear combination coefficient.

Evaluation unit 115 obtains the evaluation for assessment expression formula candidate and linear combination coefficient candidate about the assessment expression formula candidate's that obtained by combination coefficient computing unit 114 combination and linear combination coefficient candidate.That is to say that these assessment expression formula candidates and linear combination coefficient candidate are suitable for generation forecast more and estimate matrix, then high more to this assessment expression formula candidate and linear combination coefficient candidate's evaluation.For example, computing information amount benchmark is as the evaluation index for assessment expression formula candidate and linear combination coefficient candidate.

Selected cell 116 selects to have the assessment expression formula candidate's of high praise combination and linear combination coefficient candidate in assessment expression formula candidate's combination and linear combination coefficient candidate, and these are used as follow-on assessment expression formula and linear combination coefficient.

Assessment expression formula selected cell 94 will be provided to candidate's generation unit 91 and assessment expression formula holding unit 24 by follow-on assessment expression formula and the linear combination coefficient that selected cell 116 is selected.Here, the quantity K of follow-on assessment expression formula needn't be specific fixed qty, and therefore, the quantity K in each generation can fluctuate between upper limit Kmax 1.

In addition, (wherein (wherein during the evaluation of estimate of 1≤m≤M), the evaluation of estimate of input is provided to input from interface treatment facility 11 and accepts unit 21 the operation-interface treatment facility 11 of 1≤n≤N) for predetermined content Cm with input as the user Un that registers in content recommendation system in advance.Afterwards, start by recommendation apparatus 12 and be used to upgrade user that the user estimates matrix and estimate matrix update and handle.Hereinafter with reference to the flow chart among Fig. 6 the user who is undertaken by recommendation apparatus 12 being estimated the matrix update processing describes.

In step S11, the evaluation of estimate that unit 21 obtains by user's input is accepted in input.Subsequently, input is accepted unit 21 and the evaluation of estimate that is obtained is provided to the user is estimated matrix holding unit 22.

In step S12, the user estimates matrix holding unit 22 and upgrades the user who is kept based on the evaluation of estimate of accepting unit 21 from input and providing and estimate matrix, and the user estimates the matrix update processing and finishes.For example, estimating matrix holding unit 22 the user keeps the user shown in Fig. 1 to estimate matrix, and under accepting situation that unit 21 provides for the evaluation of estimate of the content C3 of user U1 from input, the user estimates matrix holding unit 22 and the evaluation of estimate that is provided is write by the user who is kept estimates in the hurdle that user U1 in the matrix and content C3 determine, upgrades the user and estimates matrix.

Like this, recommendation apparatus 12 upgrades the user who is kept and estimates matrix.Thereby when importing evaluation of estimate by the user, the user estimates matrix and is updated at every turn, can obtain more reliable prediction and evaluation matrix thus.

For prediction and evaluation value computing unit 25, by using the calculating that the user who is updated in appropriate circumstances thus estimates matrix and carries out the prediction and evaluation value from the assessment expression formula and the linear combination coefficient of assessment expression formula holding unit 24.

In addition, the assessment expression formula that the prediction and evaluation value is calculated in for example shown in Figure 7 being used to is configured to have by rudimentary algorithm carries out the operator that calculates, and promptly is used to assess the combination for the operator of the evaluation of user's content.Equally, at least one operator that comprises the calculating that is used to the carry out collaborative filtering calculating of collaborative filtering (promptly according to) in assessment in the expression formula.

Each part W11 to W14 of assessment expression formula shown in Fig. 7 comprises an operator, and the assessment expression formula is configured to the combination of these operators.For example, part W11 comprises processing axle parameter " U# " and operator " Normalize Avg ".

Here, handle the information that the axle parameter means the pending evaluation of estimate in the evaluation of estimate (Shu Ru evaluation of estimate and not any one evaluation of estimate at least of the evaluation of estimate of input before) that is used for definite user and estimates matrix (interim prediction matrix).That is to say that the evaluation of estimate of handling the identical user of axle parameter " U# " expression is used as processing target.For example, the evaluation of estimate for the content C1 to CM of user U1 becomes processing target.Equally, operator " Normalize Avg " expression user estimate evaluation of estimate in the matrix with the mean value of evaluation of estimate by normalization.

The part W12 of assessment expression formula comprises processing axle parameter " C# " and operator " FillAvg ", the evaluation of estimate of wherein handling axle parameter " C# " expression identical content is used as processing target, operator " FillAvg " expression user estimates the blank column of matrix (interim prediction matrix), and promptly the hurdle of the evaluation of estimate of not importing is received in the mean value of evaluation of estimate.

Equally, the part W13 of assessment expression formula comprises a processing axle parameter " U; C# " and operator close " CF-Pearson (8) ", wherein handle an axle parameter " U; C# " be illustrated in evaluation of estimate with identical user as after the processing target; further will be for the evaluation of estimate of identical content as processing target, operator " CF-Pearson (8) " is used for calculating the prediction and evaluation value by the evaluation of estimate with eight users of the descending of Pearson came correlation (Pearson correlation).Application is the typical case of collaborative filtering by the calculating of the Pearson came correlation that this operator is carried out.

Notice that numeral " 8 " expression in the operator " CF-Pearson (8) " is used for representing the parameter of value of the variable of the calculation process of being undertaken by operator.That is to say that, eight users' evaluation of estimate has been used in parameter " 8 " expression here.

In addition, the part W14 of assessment expression formula comprises operator " Sqrt ", and this operator " Sqrt " expression obtains the square root of each evaluation of estimate.

Thereby, for the assessment expression formula that constitutes by a plurality of operators, carry out calculating with the order from the left side operator to the right operator in the accompanying drawing.That is to say, according to the assessment expression formula shown in Fig. 7, at first, the user estimate matrix (interim prediction matrix) before the evaluation of estimate of input for each user by normalization, and the mean value of the evaluation of estimate of each content is used as the not evaluation of estimate of input of this content.Subsequently, obtain the Pearson came correlation between the user, and for each user, from this user's evaluation of estimate and eight users with high Pearson came correlation, obtain this user's the not evaluation of estimate of input, and final, obtain the square root of each evaluation of estimate, and with the value that obtained evaluation of estimate or the prediction and evaluation value as input before final.

Thereby, in some cases, according to be included in the operator of assessment in the expression formula estimate in the matrix (interim prediction matrix) by the calculation process of this operator to the user before the evaluation of estimate of input handle.

Equally, for example, for being included in the operator of assessment in the expression formula, the operator shown in can rendering 8.In Fig. 8, operator is presented in the operator denominational in left side, and is presented in the contents processing hurdle on right side by the contents processing of the calculating of the operator representation of left-hand column.

Particularly, operator " NormalizeMaxMin " is to be used for maximum and minimum value assessed value being carried out normalized operator, and operator " NormalizeAvg " is to be used to utilize the mean value of evaluation of estimate that evaluation of estimate is carried out normalized operator.

Equally, operator " FillMax " is to be used for inserting the operator that the user estimates the blank column of matrix with the maximum of evaluation of estimate, and operator " FillMin " is to be used for inserting the operator that the user estimates the blank column of matrix with the minimum value of evaluation of estimate.In addition, operator " FillMedian " is to be used for inserting the operator that the user estimates the blank column of matrix with the intermediate value of evaluation of estimate, and operator " FillAvg " is to be used for inserting the operator that the user estimates the blank column of matrix with the mean value of evaluation of estimate.

In addition, operator " CF-Correl (num) " is to extract the operator of the mean value of num evaluation of estimate as the prediction and evaluation value according to cosine distance from high correlation, and operator " CF-Pearson (num) " is the operator of the mean value of num evaluation of estimate of extraction from the highest Pearson came correlation as the prediction and evaluation value.In addition, similarly, operator " CF-Euclid (num) " is to extract the operator of the mean value of num evaluation of estimate as the prediction and evaluation value from nearest Euclidean distance.Here, the expression of " num " among operator " CF-Correl (num) ", " CF-Pearson (num) " and " CF-Euclid ian (num) " parameter.

Similarly, there is the operator that is used to carry out about the calculating of exponential function, for example is used to calculate the operator " log " of logarithm, be used for the operator " Exp " of gauge index function, be used to calculate subduplicate operator " Sqrt " or the like.In addition, also there is the operator of the calculating be used to carry out trigonometric function, for example is used to calculate the operator " Sin " of SIN function, be used to calculate the operator " Cos " of cosine function, be used to calculate operator " Tan " of tan or the like.

We say, for being used to insert the calculating of operator that the user estimates the blank column of matrix (interim prediction matrix Tk or assessment result Hs), promptly for the calculating of the operator of the prediction and evaluation value of the evaluation of estimate that is used to obtain not import, the evaluation of estimate on hurdle to be calculated is not used to the calculating of the prediction and evaluation value that obtains this hurdle.

For example, there is the assessment expression formula as shown in Fig. 9 A to 9C in another example as the assessment expression formula that can obtain by the combination of this operator.

For the assessment expression formula shown in Fig. 9 A, at first, be normalized to 0 to 1 value for each user in the evaluation of estimate that the user estimates matrix (interim prediction matrix) after, each evaluation of estimate is carried out the calculating of SIN function, obtain the cosine distance between each user thus.Subsequently, in the user of evaluation content Cm, a user who has high correlation according to the cosine distance with pending user is used as prediction and evaluation value for this pending user's content Cm for the evaluation of estimate of content Cm.

Equally, for the assessment expression formula shown in Fig. 9 B, in the content that the user has estimated, selected to have three contents of high correlation according to Euclidean distance with pending content, and will be for the mean value of the user's of selected content the evaluation of estimate prediction and evaluation value as pending content.

In addition, for the assessment expression formula shown in Fig. 9 C, the user is estimated each evaluation of estimate of matrix (interim prediction matrix) and carry out the calculating of logarithmic function, for each content, acquisition is for the mean value of each user's of this content evaluation of estimate, and the mean value that is obtained is used as the prediction and evaluation value of this content.

Recommendation apparatus 12 uses above-mentioned assessment expression formula and linear combination coefficient obtains the prediction and evaluation value that the user estimates the evaluation of estimate of not importing in the matrix, and recommendation apparatus 12 can be to each user's content recommendation thus.

For example, recommendation apparatus 12 is periodically or according to carrying out recommendation process from the instruction of interface treatment facility 11 with the indicate recommended user's of giving the recommendation information of content of generation.The recommendation process of being undertaken by recommendation apparatus 12 will describe with reference to the flow chart among Figure 10 following.

In step S41, prediction and evaluation value computing unit 25 is assessed expression argument k and is set to 1, the assessment expression formula Fk that this assessment expression argument k is used for determining providing from assessment expression formula holding unit 24 (1≤k≤K) wherein.That is to say, when the parameter k of assessment is set to 1, determine assessment expression formula F1 by assessment expression argument k=1, and this assessment expression formula F1 is used to handle.

In step S42, the evaluation of estimate that the user that interim predicting unit 51 usefulness are estimated matrix holding unit 22 and provided from the user estimates matrix is replaced and is represented the interim prediction matrix Tk of user U1 to UN for the evaluation of estimate of content.That is to say, estimate the identical information of matrix (matrix) with the user and be used as interim prediction matrix Tk for assessment expression formula Fk (1≤k≤K) wherein.For example, under the situation of assessment expression argument k=1, estimate the identical information of matrix with the user and be used as for the interim prediction matrix T1 that assesses expression formula F1.

In step S43, interim predicting unit 51 selects to constitute the operator of assessment expression formula Fk, the assessment expression formula that the selected conduct of this assessment expression formula Fk is used to handle.For example, be provided as in the assessment expression formula shown in Fig. 7 under the situation of assessment expression formula Fk, in Fig. 7, operator is sequentially selected by operator to the operator on right side from the left side.Therefore, for example, for the assessment expression formula among Fig. 7, first is selected for operator " NormalizeAvg ".

In step S44, interim 51 couples of interim prediction matrix Tk of predicting unit carry out the calculation process by selected operator representation.

In step S45, the assessment expression formula Fk that interim predicting unit 51 is determined about being used to handle, promptly whether the assessment expression formula Fk that is determined by assessment expression argument k carries out the calculation process of being carried out by all operators that constitute assessment expression formula Fk.

Determining in step S45 does not have to carry out under the situation of the calculation process of being undertaken by all operators, promptly under the situation of not selecting all operators, handles and turns back to step S43, and repeat above-mentioned processing.Particularly, select to constitute the next operator of assessment expression formula Fk, and carry out the calculation process of being undertaken by this operator.

On the other hand, in step S45, determine to have carried out under the situation of the calculation process of being undertaken by all operators, calculating for interim prediction matrix Tk is carried out by the assessment expression formula Fk that application is used to handle, and interim thus predicting unit 51 provides the interim prediction matrix Tk that carried out calculation process by using assessment expression formula Fk to predicting unit 52.Subsequently, processing proceeds to step S46.

Particularly, for example, selected the assessment expression formula Fk among Fig. 7, this means that then four operators being included among the assessment expression formula Fk are carried out calculating for interim prediction matrix Tk by selective sequential if we say.

More specifically, provide the user among Fig. 1 to estimate matrix, and selected the assessment expression formula Fk shown in Fig. 7, shown in left side among Figure 11, estimated the identical information of matrix with user among Fig. 1 and be used as interim prediction matrix Tk if we say.Subsequently, interim prediction matrix Tk calculates by the operator that is included among Fig. 7 among the part W11, and for each user, evaluation of estimate made about each user by normalization, becomes 1 for the mean value of the evaluation of estimate of each content of this user.Thereby, obtain the interim prediction matrix Tk as shown in the right side of Figure 11.

For example, when the user U1 among the interim prediction matrix Tk of concern before normalization, user U1 imports evaluation of estimate " 1 ", " 2 " and " 4 " respectively for content C1, C2 and C4.Afterwards, interim predicting unit 51 makes the mean value for the evaluation of estimate of these contents become 1 the evaluation of estimate normalization of content C1, C2 and C4.

Thereby, for example among Figure 11 shown in the right side, after calculating, obtain evaluation of estimate " 0.43 " by operator

(0.43 &cong; 1 \times 3 / (1 + 2 + 4))

As the evaluation of estimate of user U1 for content C1.

Thereby, when the calculating of the operator among the part W11 that has carried out the assessment expression formula Fk that is included among Fig. 7, afterwards, select to be included in the operator among the part W12, and as shown in figure 12, interim prediction matrix Tk calculates by this operator.

That is to say, shown in left side among Figure 12,, obtain thus at the interim prediction matrix Tk shown in the right side of described accompanying drawing estimating the mean value of being inserted the evaluation of estimate of corresponding contents by the blank column among the normalized interim prediction matrix Tk.For example, note the content C1 among the interim prediction matrix Tk of normalization, carry out by user U1, U4 and U6 and estimate, and its evaluation of estimate is " 0.43 ", " 0.89 " and " 0.57 ".

Therefore, interim predicting unit 51 obtains the mean value of these evaluations of estimate, and in the evaluation of estimate of the content C1 of interim prediction matrix Tk not the prediction and evaluation value of the evaluation of estimate of input be set to the mean value " 0.63 " that obtained

(0.63 &cong; (0.43 + 0.89 + 0.57) \div 3) .

Thereby as shown in the right side of accompanying drawing, the value column of the content C1 of the every other user except that user U1, U4 and U6 is received in prediction and evaluation value " 0.63 ".Notice that for the interim prediction matrix Tk shown in the right side of Figure 12, the hurdle for the prediction and evaluation value of the evaluation of estimate of not importing is represented to be filled with in the hurdle of shade.

When the calculating of operating part W12, further select to be included in the operator among the part W13, and interim prediction matrix Tk calculates by selected operator.That is to say that at first, about each content, calculating has been carried out for the Pearson came correlation between the user of the evaluation of content.

For example, we say and obtain the interim prediction matrix shown in Figure 13 as the result who has filled in the blank column of interim prediction matrix.Notice that in Figure 13, the hurdle for the prediction and evaluation value of the evaluation of estimate of not importing is represented to be filled with in the hurdle of shade.We say that user U2 is interested for the evaluation (prediction and evaluation value) of content C1 in 51 pairs of interim prediction matrixs of interim predicting unit, and obtain user U2 and carried out for the Pearson came correlation between another user of the evaluation of content C1.

In this case, evaluation of estimate is input to the user person of being to use U1, U4 and the U6 of content C1, so obtains the Pearson came correlation between these users and the user U2.For example, under the situation that obtains the Pearson came correlation between user U1 and the user U2, user U1 is used to acquisition Pearson came correlation for the evaluation of estimate (Shu Ru evaluation of estimate or prediction and evaluation value before) of content C2 to CM and user U2 for the evaluation of estimate of content C2 to CM.

In Figure 13, as as shown in the right side of this accompanying drawing, obtain to represent between user U2 and the U1 respectively, between user U2 and the U4 and correlation " 0.593014 ", " 0.83773 " and " 0.761491 " of the Pearson came degree of relevancy between user U2 and the U6.

For each user, do not carry out the content of evaluation about this user, when obtaining this user and having carried out for the Pearson came correlation between other users of the evaluation of described content, interim predicting unit 51 is carried out the ordering of evaluation of estimate based on the correlation that is obtained.Particularly, for each user, do not carry out the content of evaluation about this user, interim predicting unit 51 sorts with the descending of the correlation evaluation of estimate to other users.

For example, user U2 to the interested situation of the evaluation of content C1 under, as shown in figure 14, each user's evaluation of estimate sorts with the descending with the Pearson came correlation of user U2.In Figure 14, be presented at the right side of this accompanying drawing with the correlation of the Pearson came correlation of user U2.When other users' evaluation of estimate being sorted with the descending of this correlation, interim predicting unit 51 select with eight users of the descending of interested user's correlation, and obtain mean value for the evaluation of estimate of selecteed eight contents that the user paid close attention to.Subsequently, interim predicting unit 51 with the mean value that obtained as prediction and evaluation value for the content that interested user paid close attention to.

Therefore, for example,, eight user U124, U987, U25, U539, U235, U169, U206 and U83 of high correlation have been selected to have with user U2 for the example among Figure 14.Subsequently, evaluation of estimate " 0.55 ", " 0.83 ", " 0.9 ", " 1.21 ", " 0.41 ", " 0.88 " for these selected users' content C1, the mean value of " 0.52 " and " 0.46 " have been obtained.In addition, interim predicting unit 51 with the mean value " 0.72 " (0.72=(0.55+0.83+0.9+1.21+0.41+0.88+0.52+0.46)/8) that obtained as the prediction and evaluation value of user U2 for content C1.

Thereby the calculating of the part W13 of the assessment expression formula Fk in carrying out Fig. 7 and when obtaining prediction and evaluation value for evaluation content not about each user obtains the interim prediction matrix Tk shown in Figure 15.Notice that in Figure 15, the hurdle for the prediction and evaluation value of the evaluation of estimate of not importing is represented to be filled with in the hurdle of shade.

For the interim prediction matrix Tk among Figure 15, user U2 is received in the mean value " 0.72 " of the evaluation of estimate of the first eight bits user with high correlation for the value column of content C1, and this mean value obtains by the calculating of the part W13 of assessment expression formula Fk among Fig. 7.Equally, NE content bar is received in the prediction and evaluation value that obtains by the Pearson came correlation of using about each user.

When interim prediction matrix Tk had carried out the calculating of part W13 of the assessment expression formula Fk among Fig. 7, interim predicting unit 51 was further carried out the calculating of the operator shown in the part W14 of the assessment expression formula Fk among Fig. 7 to interim prediction matrix Tk.Thereby, as shown in figure 16, obtain the square root of evaluation of estimate on each hurdle of the interim prediction matrix Tk on this accompanying drawing left side, and therefore, obtain the interim prediction matrix Tk shown in this accompanying drawing right side.Notice that in Figure 16, the hurdle for the prediction and evaluation value of the evaluation of estimate of not importing is represented to be filled with in the hurdle of shade.

For example, if we see that user U1 is for the value column of content C1 in the interim prediction matrix Tk, then obtain the square root of the evaluation of estimate " 0.43 " in the interim prediction matrix Tk shown in the left side of this accompanying drawing, and the square root " 0.66 " that is obtained is inserted on the corresponding hurdle in the interim prediction matrix Tk shown in the right side of this accompanying drawing

(0.66 &cong; {(0.43)}^{1 / 2}) .

Explanation will turn back to the flow chart among Figure 10.When carrying out the calculating of all operators that constitute assessment expression formula Fk as mentioned above, in step S45, determine to have carried out the calculation process of being undertaken, and processing proceeds to step S46 by all operators.

In step S46, interim predicting unit 51 determines whether that the assessment expression formula Fk that provides about all obtains interim prediction matrix Tk, promptly whether has carried out the calculating of using corresponding assessment expression formula Fk about all interim prediction matrix Tk1.For example, be under the situation of quantity K of the assessment expression formula that provided in the quantity of the assessment expression argument that keeps by interim predicting unit 51, determine to have obtained interim prediction matrix Tk about all assessment expression formula Fk.

Definite in step S46 do not have not obtain under the situation of interim prediction matrix Tk about all assessment expression formula Fk, and interim predicting unit 51 increases the assessment expression argument k that is kept, and processing turns back to step S42 afterwards.When increasing the assessment expression argument, select the new assessment expression formula Fk that determines by the assessment expression argument, and carry out processing for interim prediction matrix Tk by using selected assessment expression formula Fk.

On the other hand, in step S46, determine to obtain under the situation of interim prediction matrix Tk, handle proceeding to step S47 about all assessment expression formula Fk.Obtaining under the situation of interim prediction matrix Tk, for example shown in Figure 17, generating corresponding K interim prediction matrix T1 to TK respectively for the K that is provided an assessment expression formula F1 to FK about all assessment expression formula Fk.Therefore, generate K interim prediction matrix altogether respectively according to K appraisal procedure.

In step S47, predicting unit 52 is used from the interim prediction matrix Tk of interim predicting unit 51 and is calculated user Un from the linear combination coefficient of assessment expression formula holding unit 24 (wherein 1≤n≤N) is for the final prediction and evaluation value of each content.

That is to say a user Un among the user U1 to UN of predicting unit 52 selection registered in advance.For example, user U1 to UN is sequentially selected.Subsequently, predicting unit 52 is extracted the interim user's prediction matrix Tnk of information conduct that is made of the evaluation of estimate for each content of selected user Un in the interim prediction matrix Tk, and wherein interim prediction matrix Tk provides from interim predicting unit 51.Thereby the K altogether of user Un interim user's prediction matrix Tnk obtains from K interim prediction matrix Tk respectively.

In addition, as shown in figure 18, (wherein 0≤k≤K) carries out linear combination to the interim user's prediction matrix Tn1 to Tnk of 52 couples of K that obtained of predicting unit by using linear combination coefficient bnk from the user Un that assesses expression formula holding unit 24.Thereby acquisition is by user's prediction matrix An of the user Un of evaluation of estimate (by the evaluation of estimate of input before the computing) formation of final prediction and evaluation value and input before.

More specifically, predicting unit 52 is by calculating user's prediction matrix An that following formula (1) obtains user Un.

An = (Σ_{k = 1}^{K} bnk \times Tnk) + bn 0 - - - (1)

Note, in expression formula (1), (expression of ∑ bnk * Tnk) by with the variable k of linear combination coefficient bnk and interim user's prediction matrix Tnk from 1 change to K obtain interim user's prediction matrix Tnk and linear combination coefficient bnk product with.Equally, linear combination coefficient bn1 to bnK is the linear combination coefficient corresponding to interim user's prediction matrix Tn1 to TnK, and linear combination coefficient bn0 is the coefficient that is added into the linear summation of interim user's prediction matrix Tnk.

Therefore, the product by linear combination coefficient bn0 being added to interim user's prediction matrix Tnk and linear combination coefficient bnk and in obtain user's prediction matrix An of user Un.For example, by give for each prediction and evaluation of the content Cm of the user Un in each interim user's prediction matrix Tnk prediction and evaluation value that obtains to multiply by linear combination coefficient bnk with each linear combination coefficient bnk on duty and, and further linear combination coefficient bn0 is added to above-mentioned and in, thereby obtain the final prediction and evaluation value of user Un for content Cm.

When having obtained user's prediction matrix An, predicting unit 52 is provided to prediction and evaluation value matrix holding unit 26 with the user's prediction matrix An that is obtained.

In step S48, predicting unit 52 determines whether to obtain user's prediction matrix An about all users.Determine not obtain under the situation of user's prediction matrix An about all users in step S48, handle and turn back to step S47, selecting in step S47 does not also have selecteed user Un, and obtains user's prediction matrix An of this user Un.

On the other hand, in step S48, determine to obtain under the situation of user's prediction matrix An, handle proceeding to step S49 about all users.When from predicting unit 52 when prediction and evaluation matrix holding unit 26 provides user's prediction matrix A1 to AN about all user U1 to UN, a prediction and evaluation matrix A that is made of the N that is provided user's prediction matrix A1 to AN is provided prediction and evaluation matrix holding unit 26.Subsequently, prediction and evaluation matrix holding unit 26 keeps the prediction and evaluation matrix A that generated, and this prediction and evaluation matrix A is provided to recommendation unit 27.

In step S49, to user's content recommendation, and recommendation process finishes recommendation unit 27 based on the prediction and evaluation matrix that provides from prediction and evaluation matrix holding unit 26.For example, recommendation unit 27 generates recommendation information, and this recommendation information is represented the content about each user Un recommendation, and the content of this user Un is recommended in the content conduct that wherein has the maximum prefetch test and appraisal value of user Un.Subsequently, recommendation unit 27 is provided to interface treatment facility 11 to show this recommendation information with the recommendation information that generates.Like this, the user watches the recommendation information that is shown, and the user can seek and may make the interested new content of user thus.

As mentioned above, recommendation apparatus 12 is used a plurality of assessment expression formulas and is generated the interim prediction matrix of assessing expression formulas corresponding to these, and by linear combination coefficient interim prediction matrix is carried out linear combination, generates final prediction matrix thus.

Therefore, a plurality of interim prediction matrix for a plurality of assessment expression formulas carries out linear combination by linear combination coefficient, thereby when considering each recommendation results that obtains by polyalgorithm, make it possible to be assessed as the favorite content of user with the higher accuracy detection.

For example, about predetermined user Un, we say in expectation and detect under the situation that is assessed as the favorite content of user Un, exist to be used to extract by the assessment expression formula F1 of the content of all user's high evaluations and to be used to extract and had the assessment expression formula F2 of content of user's high evaluation of high correlation with user Un.In this case, when detecting the content that will recommend user Un by independent application assessment expression formula F1, even thinking very highly of providing of all users, but also not necessarily user Un is favorite by detecting the content that obtains.

On the contrary, when detecting the content that to recommend user Un by independent application assessment expression formula F2, carry out the detection of content, wherein do not consider all users' hobby but consider part user's hobby, therefore not necessarily user Un is favorite by detecting the content that obtains.

On the other hand, recommendation apparatus 12 is used by unit 23 optimized a plurality of assessment expression formulas and is obtained final assessment result by the assessment result of these assessment expression formulas is carried out linear combination.Therefore, considering for example all users, having under the situation of a plurality of appraisal procedures such as user of high correlation with user Un and obtain final assessment result (prediction and evaluation matrix), can obtain be assessed as the favorite content of user Un thus with higher accuracy.

In addition, carry out the study of the assessment expression formula of using last generation by unit 23 and handle, order generates and is used for assessment expression formula and the linear combination coefficient that generation forecast is estimated each generation of matrix.Handle and to describe with reference to the flow chart among Figure 19 following by the study that unit 23 is carried out.

In step S81,91 application of candidate's generation unit are carried out assessment expression formula candidate from K the assessment expression formula in the last generation that assessment expression formula selected cell 94 provides and are generated processing, generate S assessment expression formula candidate Js thus, this S assessment expression formula candidate Js is the candidate of follow-on assessment expression formula.Subsequently, candidate's generation unit 91 assessment expression formula candidate Js that will generate is provided to assessment result computing unit 92.

In step S82, assessment result computing unit 92 calculates each assessment expression formula candidate's assessment result Hs thus by using from the assessment expression formula candidate Js of candidate's generation unit 91 and estimating matrix from the user that the user estimates matrix holding unit 22 and carry out evaluation process and assess the user and estimate corresponding evaluation of estimate in the matrix.Assessment result computing unit 92 is provided to assessment expression formula selected cell 94 with assessment result Hs and the assessment expression formula candidate Js that calculates.

Notice that the back will illustrate the details that generates processing and evaluation process about assessment expression formula candidate.

In step S83, group generation unit 93 is estimated matrix based on the user who estimates matrix holding unit 22 from the user and provide and is carried out user's cluster (clustering).Particularly, group generation unit 93 user's brief introduction of providing in appropriate circumstances is provided among the registered user U1 to UN each is categorized into a predetermined Q user is organized among the G1 to GQ any one group.

For example, group generation unit 93 is carried out the grouping of user U1 to UN based on estimating each user in the matrix the user who is provided for the evaluation of estimate of content, carry out cluster by K average (K-means) method.Thereby as shown in figure 20, each user is classified into the user and organizes in one of any user group among the G1 to GQ.For example, in Figure 20, the execution grouping makes user U1, U3, U10 and U152 belong to the user and organizes G1.

Explanation will turn back to the flow chart among Figure 19.When carrying out user's grouping, group generation unit 93 will represent that user's group information of its group result is provided to assessment expression formula selected cell 94.

In step S84, assessment expression formula selected cell 94 use the user who estimates matrix holding unit 22 from the user estimate matrix, from the assessment result Hs and the assessment expression formula candidate Js of assessment result computing unit 92 and carry out the assessment expression formula from user's group information of group generation unit 93 and select to handle, generate follow-on assessment expression formula and linear combination coefficient thus.Subsequently, the assessment expression formula selected cell 94 assessment expression formula and the linear combination coefficient that will generate is provided to candidate's generation unit 91 and assesses expression formula holding unit 24.Assessment expression formula holding unit 24 keeps assessment expression formula and the linear combination coefficient from assessment expression formula selected cell 94.Notice that the details that the assessment expression formula is selected to handle will be described hereinafter.

In step S85, unit 23 determines whether to finish to be used to generate the processing of follow-on assessment expression formula and linear combination coefficient.For example, under the situation of having indicated the operation of ending content recommendation system, determine end process.

In step S85, determine under the situation of end process not, handle turning back to the step S81 that generates follow-on assessment expression formula and linear combination coefficient.On the other hand, in step S85, determine under the situation of end process, the processing that each unit in the unit 23 stops to be performed, and the study processing finishes.

Like this, the genetic programming of the assessment expression formula of unit 23 by using last generation generates follow-on assessment expression formula and linear combination coefficient.Thereby follow-on assessment expression formula and linear combination coefficient generate by genetic programming, make it possible to obtain the more appropriate prediction and evaluation value of user for each content thus.

In addition, for unit 23, can under the situation of the keeper's who does not have recommendation apparatus 12 etc. instruction, set up and be used for easier and more promptly to the proposed algorithm of user's content recommendation.That is to say, can obtain the assessment expression formula of evaluation prediction evaluation of estimate more accurately, and the combination and the linear combination coefficient of the assessment expression formula that can obtain to be more suitable for.

Afterwards, generating processing with reference to the flow chart among Figure 21 about the assessment expression formula candidate corresponding to the processing among the step S81 among Figure 19 describes.

In step S121, candidate's generation unit 91 determines whether this study is study for the first time.Particularly, when carrying out study, generate assessment expression formula and linear combination coefficient, and will assess expression formula and be provided to candidate's generation unit 91 from assessing expression formula selected cell 94.On the other hand, be under the situation of study for the first time fully in this study, both do not generate the assessment expression formula and do not generated linear combination coefficient yet, therefore not assessing expression formula is provided to candidate's generation unit 91.Therefore, do not providing under the situation of assessment expression formula from assessment expression formula selected cell 94, candidate's generation unit 91 determines that this study is study for the first time.

Determine that in step S121 this study is under the situation of study for the first time, in step S122, candidate's generation unit 91 selects quantity to be set to NumSlct=0, the quantity of suddenling change is set to NumMts=0, number of crossovers is set to NumC rs=0, and generates quantity at random and be set to NumRnd=S.

Here, select quantity NumSlct to be illustrated in the quantity of selected assessment expression formula as follow-on assessment expression formula Js in the assessment expression formula in last generation.Equally, sudden change quantity NumMts, number of crossovers NumCrs and generate the quantity that quantity NumRnd is illustrated respectively in the candidate Js of the assessment that generates in sudden change processing, cross processing and the random process at random.

The assessment expression formula that does not have last generation when learning for the first time, therefore generating quantity at random is set to S, and all S assessment expression formula candidates generate in random process.After having determined selection quantity, sudden change quantity, number of crossovers and having generated quantity at random, handle proceeding to step S124.

On the other hand, determine that in step S121 this study is not under the situation of study for the first time, in step S123, candidate's generation unit 91 is provided with selects quantity NumSlct=K, sudden change quantity NumMts=(S-K)/3, number of crossovers NumCrs=(S-K)/3, and generate quantity NumRnd=S-(2 * (S-K)/3)-K at random.

That is to say that from K the assessment expression formula in the last generation that assessment expression formula selected cell 94 provides, promptly all assessment expression formulas in last generation are set to follow-on assessment expression formula candidate.Simultaneously, according to sudden change processing, cross processing and random process, the quantity that assessment expression formula candidate Js generates is to account for by deduct 1/3 of quantity that K obtains from corresponding assessment expression formula candidate's summation S.After having determined selection quantity, sudden change quantity, number of crossovers and having generated quantity at random, handle proceeding to step S124.

When in step S122 or S123, having determined selection quantity, sudden change quantity, number of crossovers and having generated quantity at random, in step S124, candidate's generation unit 91 is selected assessment expression formula candidate from the assessment expression formula in last generation of being provided by assessment expression formula selected cell 94, wherein the quantity of the Xuan Zeing quantity that the quantity NumSlct of selection represents of serving as reasons.For example, under the situation of selecting quantity NumSlct=K, the assessment expression formula F1 to FK in all K that provide last generations is used as assessment expression formula candidate J1 to JK.

In step S125, sudden change processing unit 111 the assessment expression formula that provides from assessment expression formula selected cell 94 is provided is carried out sudden change and handle, and generates the serve as reasons assessment expression formula candidate Js of the quantity that sudden change quantity NumMts represents of quantity thus.

That is to say, at random select an assessment expression formula among the assessment expression formula F1 to FK of 111 the past of sudden change processing unit generation.Subsequently, sudden change processing unit 111 pairs of selected assessment expression formulas are carried out the conversion etc. of order of modification, the operator of the insertion of operator, the removal of operator, the modification of handling the axle parameter, the parameter in the operator, generate an assessment expression formula candidate thus.

For example, we say that the assessment expression formula of a generation has been selected the assessment expression formula shown in Figure 22 A in the past.Assessment expression formula shown in Figure 22 A is made of three part W31 to W33, and part W31 to W33 comprises operator " NormalizeMaxMin ", " Sin " and " CF-Correl (1) " respectively.

For example, when passing through new processing axle parameter of sudden change processing increase and operator between part W32 in the assessment expression formula in Figure 22 A and the W33, generate the assessment expression formula candidate shown in Figure 22 B.For the assessment expression formula candidate shown in Figure 22 B, new processing axle parameter " C# " and operator " FillAvg " are added between the part W32 and W33 in the assessment expression formula among Figure 22 A.

Equally, for example, when handling the part of removing the assessment expression formula among Figure 22 A, generate the assessment expression formula candidate shown in Figure 22 C by sudden change.For the assessment expression formula candidate shown in Figure 22 C, the part W31 of the assessment expression formula among Figure 22 A is removed.

In addition, when the processing axle parameter of the part W31 in the assessment expression formula among Figure 22 A is modified, generate the assessment expression formula candidate shown in Figure 22 D.For the assessment expression formula candidate shown in Figure 22 D, the processing axle parameter " U# " of the part W31 of assessment expression formula is changed and is " C# " among Figure 22 A.

In addition, handling by sudden change after the assessment expression formula make among Figure 22 A changes into state among Figure 22 D, further when the parameter in the operator " CF-Correl (1) " in the assessment expression formula is modified, for example generate the assessment expression formula candidate shown in Figure 22 E.For the assessment expression formula candidate shown in Figure 22 E, the parameter of the right-hand member operator in the assessment expression formula among Figure 22 D " CF-Correl (1) " is changed and is " CF-Correl (5) ".

In addition, similarly,, further when the order of operator is transformed, for example generate the assessment expression formula candidate shown in Figure 22 F handling by sudden change after the assessment expression formula make among Figure 22 A changes into state among Figure 22 E.For the assessment expression formula candidate shown in Figure 22 F, the order between the operator " Sin " of the assessment expression formula among Figure 22 E and the operator " CF-Correl (5) " is transformed to " U; C#CF-Correl (5), Sin ".

Therefore like this, handle a feasible part of assessing expression formula by sudden change and change into new assessment expression formula candidate.Note, estimate under the situation that matrix calculates by the assessment expression formula candidate who use to generate the user, when generating the assessment expression formula candidate that blank column that the user estimates matrix do not fill in, promptly when generating the assessment expression formula candidate not have to obtain the prediction and evaluation value of the evaluation of estimate of input not, do not use this assessment expression formula candidate and regenerate and assess the expression formula candidate.

Explanation will turn back to the flow chart among Figure 21.When handling generation assessment expression formula candidate by sudden change, in step S126,112 application of cross processing unit are carried out cross processing from the assessment expression formula that assessment expression formula selected cell 94 provides, generate assessment expression formula candidate Js thus, the quantity of the quantity of this assessment expression formula candidate Js for representing by number of crossovers NumCrs.

That is to say, select two assessment expression formulas arbitrarily among the assessment expression formula F1 to FK of 112 the past of cross processing unit generation.Subsequently, the part of selected two the assessment expression formulas of cross processing unit 112 conversions generates a new assessment expression formula candidate thus.

For example, as shown in figure 23, we say and have selected assessment expression formula that is made of part W51 and W52 and the assessment expression formula that is made of part W53 and W54.Here, part W52 comprises two operators " Sin " and " CF-Correl (1) ", and part W53 comprises an operator " CF-Correl (3) ".Subsequently, when the part W51 of the assessment expression formula that is made of part W51 and W52 is converted (change) for the part W53 of another assessment expression formula, generate the new assessment expression formula candidate " C shown in the downside be presented at accompanying drawing; U#CF-Pearson (3), Sin, U; C#CF-Correl (1) ".

Like this, the part of assessment expression formula is converted to the part of another assessment expression formula by cross processing, and is used as new assessment expression formula.Note, estimate under the situation that matrix calculates by the assessment expression formula candidate who use to generate the user, when generating the assessment expression formula candidate that blank column that the user estimates matrix do not fill in, do not use this assessment expression formula candidate, and regenerate and assess the expression formula candidate.

Explanation will turn back to the flow chart among Figure 21.When being generated assessment expression formula candidate by cross processing, in step S127, random process is carried out in random process unit 113, generates assessment expression formula candidate Js thus, and the quantity of this assessment expression formula candidate Js is served as reasons and generated the quantity that quantity NumRnd represents at random.

That is to say that random process unit 113 will be handled a parameter randomly arbitrarily and combine with the operator of arbitrary parameter, generate assessment expression formula candidate.In this case, also determine the quantity and the order of operator randomly.Note, estimate under the situation that matrix calculates by the assessment expression formula candidate who use to generate the user, when generating the assessment expression formula candidate that blank column that the user estimates matrix do not fill in, do not use this assessment expression formula candidate, and regenerate and assess the expression formula candidate.

When selecting quantity, sudden change quantity, number of crossovers and generating quantity at random to generate assessment expression formula candidate, the assessment expression formula candidate J1 to JS of this generation is provided to assessment result computing unit 92 from candidate's generation unit 91, and assessment expression formula candidate generates processing and finishes, and handles the step S82 that proceeds among Figure 19.

Like this, candidate's generation unit 91 generates follow-on assessment expression formula candidate by the assessment expression formula of using last generation.Thereby the follow-on assessment expression formula candidate assessment expression formula of a generation in the past generates, thus in several combinations of having estimated assessment expression formula candidate with assessed under the situation of user for the evaluation of content the assessment expression formula that can obtain to be more suitable for.

Below, with reference to the flow chart among Figure 24 the evaluation process corresponding to the processing among the step S82 of Figure 19 is described.

In step S151, the assessment expression formula candidate that assessment result computing unit 92 selects a conduct from S the assessment expression formula candidate J1 to JS that candidate's generation unit 91 provides to be applied to handle.For example, the assessment expression formula candidate Js that is applied to handle (1≤s≤S) sequentially from assessment expression formula candidate J1 to JS, select wherein.

In step S152, assessment result computing unit 92 will be estimated the identical matrix of matrix (information) as the assessment result Hs for selected assessment expression formula candidate Js with the user who estimates matrix holding unit 22 from the user and provide, and the evaluation of estimate of estimating each hurdle of matrix with the user is replaced each hurdle of assessment result Hs.That is to say, estimate the identical information of matrix with the user and be used as assessment result Hs.

In step S153, the operator of the assessment expression formula candidate Js that 92 selections of assessment result computing unit are used to handle.For example, be used as the assessment expression formula candidate shown in Figure 25 under the situation of assessment expression formula candidate Js, in Figure 25, sequentially select from left side operator to right side operator with direction to the right.In Figure 25, assessment expression formula candidate Js comprises four operators " CF-Pearson (3) ", " Sin ", " CF-Correl (1) " and " Cos " with following this selective sequential.

In step S154, assessment result computing unit 92 determines whether selected operator is the last operator that will insert the blank column of assessment result Hs.

For example, the assessment expression formula candidate Js shown in Figure 25 comprises four operators.In these operators, operator " CF-Pearson (3) " and " CF-Correl (1) " are respectively by using the Pearson came correlation and inserting the operator of the blank column of assessment result Hs according to the correlation of cosine distance, promptly are the operators that is used to the prediction and evaluation value of the evaluation of estimate that obtains not import.On the other hand, operator " Sin " and " Cos " are used to obtain the sine of evaluation of estimate and the operator of cosine, and are not the operators of inserting the blank column of assessment result Hs.

Equally, for assessment expression formula candidate Js, operator " CF-Pearson (3) ", " Sin ", " CF-Correl (1) " and " Cos " are selected and be used for processing with above this order.Therefore, for assessment expression formula candidate Js, the last operator of filling in the blank column of assessment result Hs is operator " CF-Correl (1) ", and under the situation of selecting this operator, determines that in step S154 this operator is last operator.

Determine that in step S154 selected operator is not that in step S155,92 couples of assessment result Hs of assessment result computing unit carry out the calculating by selected operator representation under the situation of last operator.

Here, under selected operator is the situation of evaluation of estimate as the operator of target with all hurdles of assessment result Hs, for example be used to calculate the operator " Sin " of sine of the evaluation of estimate on each hurdle, the evaluation of estimate on each hurdle of 92 couples of assessment result Hs of assessment result computing unit is carried out the calculating by selected operator representation.Equally, at selected operator is to fill under the situation of operator of blank column of assessment result Hs, assessment result computing unit 92 is used the evaluation of estimate on each hurdle of assessment result Hs, promptly before the evaluation of estimate of input and the prediction and evaluation value of the evaluation of estimate that the evaluation of estimate do not imported obtains not import.

After the calculating of carrying out selected operator, handle and proceed to step S157 from step S155.

On the other hand, determine that in step S154 selected operator is that in step S156, all evaluations of estimate of 92 couples of assessment result Hs of assessment result computing unit are carried out the calculating by selected operator representation under the situation of last operator.

Particularly, assessment result computing unit 92 is calculating by last operator when filling in the hurdle, in the evaluation of estimate on each hurdle of assessment result Hs, not only obtains the not evaluation of estimate of input, and the prediction and evaluation value of the evaluation of estimate of input before obtaining.Note, in this case,, the evaluation of estimate on calculated hurdle to be not used in the calculating of the prediction and evaluation value that obtains this hurdle for the calculating of the operator of the blank column that is used to fill in assessment result Hs.

Therefore, also the hurdle about the evaluation of estimate of input before being transfused to obtains the prediction and evaluation value when the calculating on the hurdle of filling in by last operator, thus can by use the prediction and evaluation value that obtains about this hurdle and user estimate corresponding hurdle in the matrix before the evaluation of estimate of input estimate the accuracy of assessment expression formula candidate's prediction (assessment).After in step S156, carrying out the calculating of selected operator, subsequently, handle proceeding to step S157.

In step S157, assessment result computing unit 92 determines whether the selected operator of assessment expression formula candidate Js is the last operator that is included among the assessment expression formula candidate Js.For example, under the situation of having selected the assessment expression formula candidate Js shown in Figure 25, when the selected operator of assessment expression formula candidate Js was the operator " Cos " of the last calculating of carrying out, selected operator was confirmed as last operator.

Determine that in step S157 selected operator is not under the situation of last operator, handle and turn back to step S153, select next operator at this, and carry out calculating for assessment result Hs.

On the other hand, determine that in step S157 selected operator is that in step S158, assessment result computing unit 92 determines whether that the assessment expression formula candidate Js that provides about all obtains assessment result Hs under the situation of last operator.

For example, obtain under the situation of assessment result Hs, as shown in figure 26,, obtain S corresponding assessment result H1 to HS respectively for the S that is provided an assessment expression formula candidate J1 to JS at the assessment expression formula candidate Js that determines to provide about all.That is to say,, generate S assessment result altogether respectively according to S appraisal procedure.

In step S158, determine not obtain to handle and turn back to step S151 under the situation of assessment result Hs, repeat above-mentioned processing herein about all assessment expression formula candidate Js.Particularly, select next assessment expression formula candidate Js, and generate assessment result Hs corresponding to this assessment expression formula candidate Js.

On the other hand, determine to have obtained under the situation of assessment result Hs about all assessment expression formula candidate Js in step S158, assessment result computing unit 92 is provided to assessment expression formula selected cell 94 with S the assessment result H1 to HS that is obtained with assessment expression formula candidate J1 to JS.Subsequently, when from assessment result computing unit 92 after assessment expression formula selected cell 94 provides assessment expression formula candidate and assessment result, evaluation process finishes, and handles the step S83 that proceeds among Figure 19.

Like this, assessment result computing unit 92 is used S assessment expression formula candidate respectively and is generated corresponding assessment result.Thereby, use assessment expression formula candidate and generate corresponding assessment result, thus in the combination of having estimated assessment expression formula candidate with assessed under the situation of user for the evaluation of content, can obtain to assess the combination that is more suitable for of expression formula.

Below, describe selecting to handle with reference to the flow chart among Figure 27 corresponding to the assessment expression formula of the processing among the step S84 of Figure 19.

In step S191, the operating position of each assessment result Hs of providing from assessment result computing unit 92 is provided for combination coefficient computing unit 114 initializing variable Z, this variable Z.

That is to say,, supposed to select S some that assess among the expression formula candidate, and selected assessment expression formula candidate Js is used as follow-on assessment expression formula for assessment expression formula selected cell 94.Subsequently, be used to obtain for selected assessment expression formula candidate Js estimate this prediction and evaluation value, i.e. the accuracy of the prediction of assessment result Hs under each user's the situation of prediction and evaluation value.

Variable Z is made of the information of expression corresponding to the operating position of the assessment result Hs of each assessment expression formula candidate Js, and this operating position represents whether be used as follow-on assessment expression formula corresponding to the assessment expression formula candidate of assessment result temporarily.That is to say that for variable Z, the assessment expression formula candidate who is set to the assessment result of " using " corresponding to operating position is used as the assessment expression formula candidate of follow-on assessment expression formula temporarily.Equally, for variable Z, the assessment expression formula candidate who is set to the assessment result of " using " corresponding to operating position is the assessment expression formula candidate who temporarily is not used as follow-on assessment expression formula.

Combination coefficient computing unit 114 is set to " not using " by the operating position of each assessment result of variable Z and comes initializing variable Z.

In step S192, the assessment expression formula candidate Js that combination coefficient computing unit 114 is selected to provide from assessment result computing unit 92 (1≤s≤S) wherein.For example, assessment expression formula candidate Js is sequentially selected to assessing the expression formula candidate JS from assessment expression formula candidate J1.

In step S193, combination coefficient computing unit 114 counter-rotating is corresponding to the operating position of the assessment result Hs of selected assessment expression formula candidate Js, and the variable Z that operating position has been reversed is as variable Zs ' (1≤s≤S) wherein.For example, at variable Z is under the situation of " using " in the operating position corresponding to the assessment result Hs of selected assessment expression formula candidate Js, operating position is reversed to " using ", and on the contrary, for under the situation of " using ", operating position is reversed to " not using " in operating position.

In step S194, combination coefficient computing unit 114 is determined the quantity of operating position for " using ", promptly whether is equal to or less than the upper limit quantity Kmax of the quantity K of predetermined assessment expression formula for the assessment expression formula candidate's of the assessment result of " using " quantity corresponding to operating position.

Determine that in step S194 operating position is equal to or less than under the situation of upper limit quantity Kmax for the quantity of " using ", in step S195, combination coefficient computing unit 114 selects a user to organize Gq with reference to the user's group information that provides from group generation unit 93.For example, the user organizes Gq (wherein 1≤q≤Q) organizes G1 from the user and organizes the GQ to the user and sequentially select.

In step S196, the operating position of combination coefficient computing unit 114 choice variable Zs ' among the assessment result Hs that provides from assessment result computing unit 92 is the assessment result Hs of " using ", and with selected assessment result Hs again as assessment result Hv '.Notice that under the situation that has a plurality of new assessment result Hv ', the variable v among additional these assessment results Hv ' is so that become continuous number.

For example, we say, for variable Zs ', assessment result H1 to HV (wherein the operating position of 1≤V≤S) is " using ", and other assessment results H (V+1) to the operating position of HS be " using ".In this case, combination coefficient computing unit 114 with assessment result H1 to HV again as assessment result H1 ' to HV ', and will corresponding to assessment result H1 ' to the assessment expression formula candidate Js of HV ' again as assessment expression formula candidate J1 ' to JV '.

In step S197, the user that combination coefficient computing unit 114 uses assessment result Hv ', estimate matrix holding unit 22 from the user estimates matrix and obtains to organize about selected user the linear combination coefficient candidate Bsq of Gq from user's group information of group generation unit 93.

Here, the linear combination coefficient candidate Bsq that organizes Gq about selected user belongs to the linear combination coefficient candidate that the user organizes the user Un of Gq, and this linear combination coefficient candidate Bsq is by constituting corresponding to linear combination coefficient bnk (V+1) individual linear combination coefficient bsq0 to bsqV.That is to say that be used as to JV ' at assessment expression formula candidate J1 ' under the situation of follow-on new assessment expression formula, linear combination coefficient candidate Bsq is a linear combination coefficient.

Notice that (wherein variable s, q among 0≤v≤V) and v correspond respectively to s, user among the variable Zs ' and organize q among the Gq and the v among the assessment result Hv ' at each linear combination coefficient bsqv.Equally, linear combination coefficient bsq0 is the coefficient corresponding to linear combination coefficient bn0.

For example, combination coefficient computing unit 114 is used linear regression and is obtained linear combination coefficient candidate Bsq, and wherein the error matrix Eq that is made of the predetermined-element in the following formula (2) becomes minimum value, is the combination of linear combination coefficient bsqv.

Rq = (Σ_{v = 1}^{V} bsqv \times {Hqv}^{,}) + bsq 0 + Eq - - - (2)

In expression formula (2), matrix Rq has the user for the evaluation of estimate of input before the content matrix (vector) as element, and wherein the user belongs to the user that the user estimates in the matrix and organizes Gq.

Similarly, matrix H qv ' in the expression formula (2) is in the prediction and evaluation value with assessment result Hv ', corresponding to the user for the prediction and evaluation value of the evaluation of estimate of input before the content matrix (vector) as element, wherein the user belongs to the user that the user estimates in the matrix and organizes Gq.That is to say, the predetermined-element rq of matrix Rq is the authentic assessment value of predetermined user Un for predetermined content Cm, is that user Un by using assessment expression formula candidate Jv ' assessment is for the prediction and evaluation value of the evaluation of content Cm corresponding to the element hqv of the matrix H qv ' of this element rq.

In addition, in expression formula (2), the variable v of (∑ bsqv * Hqv ') expression linear combination coefficient bsqv and matrix H qv ' changes to V from 1, obtain thus matrix H qv ' and linear combination coefficient bsqv product and.

According to expression formula (2), the element eq of linear combination coefficient bsq0 and error matrix Eq be added into the corresponding matrix H q1 ' of element rq of matrix Rq to the element hq1 to hqV of HqV ' respectively with the summation of the product of linear combination coefficient bsq1 to bsqV in, and therefore, the result who is obtained equals element rq.That is to say that element eq represents the assessed value of element rq and the error between the true element rq, wherein obtain in the summation of the product of the assessed value of element rq by linear combination coefficient bsq0 being joined element hqv and linear combination coefficient bsqv.

Therefore, eq is more little for element, the then combination of the linear combination coefficient bsqv by using this moment and corresponding to the assessment expression formula candidate Jv ' of assessment result Hv ', and the user can assess with higher accuracy for the evaluation of content.Therefore, combination coefficient computing unit 114 obtains to make the relation of expression formula (2) to set up and elements corresponding eq becomes the combination of minimum linear combination coefficient bsqv, and with its combination as linear combination coefficient candidate Bsq.

When obtaining to handle and to proceed to step S198 after user about variable Zs ' organizes the linear combination coefficient candidate Bsq of Gq from step S197.

In step S198, combination coefficient computing unit 114 determines whether that organizing Gq about whole user obtains linear combination coefficient candidate Bsq.

In step S198, determine not organize under the situation of Gq acquisition linear combination coefficient candidate Bsq, handle and turn back to step S195, repeat above-mentioned processing at this about whole user.Particularly, select next user to organize Gq, and the linear combination coefficient candidate Bsq of Gq is organized in acquisition about the user of the new selection of variable Zs '.

On the other hand, in step S198, determine to organize under the situation of Gq acquisition linear combination coefficient Bsq, handle proceeding to step S199 about whole user.For example, determining to organize under acquisition linear combination coefficient candidate's the situation, obtain to organize the linear combination coefficient candidate Bs1 to BsQ of G1 to GQ respectively about variable Zs ' for the user about whole user.

In step S199, evaluation unit 115 is estimated matrix and assessment result Hv ' based on the user and is calculated AIC (Akaike Information Criteria about variable Zs ', Akaike's Information Criterion), assessment result Hv ' the i.e. result of calculation of the linear combination coefficient by expression formula (2) wherein.That is to say, evaluation unit 115 calculates following formula (3), calculate AIC about variable Zs ' thus as the amount of information benchmark, this amount of information benchmark is to be used to assess the evaluation index of user for the assessment expression formula candidate Jv and the linear combination coefficient candidate Bsq of the evaluation of content.

AIC＝W×{log(2×PI)+1+log‖E‖ ²/W}+2×Q×(v+1) (3)

In expression formula (3), W is illustrated in the quantity that the user estimates the evaluation of estimate of importing before in the matrix, and PI represents π.Simultaneously, in expression formula (3), ‖ E ‖ represents the norm of the matrix E that the element by error matrix E1 to EQ constitutes, the i.e. quadratic sum of each element of matrix E.In addition, in expression formula (3), Q and V represent that respectively the user organizes the quantity of Gq and the quantity of assessment result Hv.

The element of matrix E becomes more little, then just becomes more little at the AIC shown in the expression formula (3).That is to say that can accurately assess the evaluation of user for content about the variable Zs ' that assesses expression formula candidate Jv ' and linear combination coefficient candidate Bsq, then the AIC of variable Zs ' becomes more little.

Equally, the quantity (quantity of free parameter) of assessment expression formula candidate Jv ' is many more, and then the AIC of the variable Zs ' shown in the expression formula (3) becomes big more.Estimate at recommendation apparatus 12 generation forecasts under the situation of matrix, the quantity of assessment expression formula candidate Fk is many more, the amount of calculation that is used to obtain the prediction and evaluation matrix becomes big more, and the quantity of therefore assessing expression formula candidate Fk is few more, and the quantity of promptly assessing expression formula candidate Jv ' is few more good more.That is to say that from the angle of amount of calculation, the AIC of variable Zs ' is more little, the just suitable more assessment user of the assessment expression formula candidate Jv ' of variable Zs ' and linear combination coefficient candidate Bsq is for the evaluation of content.

Therefore, the AIC of variable Zs ' is more little, and just high more about the evaluation for assessment expression formula candidate Jv ' and linear combination coefficient candidate Bsq that this variable Zs ' obtains, this higher evaluation is suitable as assessment expression formula and linear combination coefficient.

In step S199, after the AIC that calculates variable Zs ', handle proceeding to step S200.

Simultaneously, in step S194, determine for variable Zs ', operating position is not equal to for the quantity of " using " and is not less than under the situation of upper limit quantity Kmax of K assessment expression formula, and then the processing among the step S195 to S199 is skipped, and processing proceeds to step S200.

That is to say, for variable Zs ', operating position is higher than under the situation of upper limit quantity Kmax for the quantity of " using ", when the assessment result that by the application operating position is " using " generated the assessment expression formula, the quantity of the assessment expression formula of generation surpassed the upper limit quantity Kmax of assessment expression formula.Therefore, be higher than under the situation of upper limit quantity Kmax in the quantity of operating position for " using ", the processing among the step S195 to S199 is skipped.

As calculating AIC in step S199 or when determining that the quantity of operating position for " using " is higher than upper limit quantity Kmax, in step S200, combination coefficient computing unit 114 determines whether to have selected all assessment expression formula candidate Js.Particularly, the processing selecting in step S192 under the situation of all S assessment expression formula candidate Js, make the judgement of selecting all assessment expression formula candidates.

In step S200, determine also not select to handle and turn back to step S192 under the situation of all S assessment expression formula candidate Js, repeat above-mentioned processing herein.Particularly, select next assessment expression formula candidate Js, and for so far variable Zs ', be inverted corresponding to the operating position of the assessment result Hs of the new assessment expression formula candidate Js that selects, it is as new variable Zs '.Subsequently,, obtain the linear combination coefficient candidate Bsq that each user organizes Gq, and calculate the AIC of variable Zs ' about variable Zs ' corresponding to selected assessment expression formula candidate Js.

On the other hand, in step S200, determined to select to handle proceeding to step S201 under the situation of all assessment expression formula candidate Js.After having selected all assessment expression formula candidate J1 to JS, this means the AIC that has obtained corresponding to variable Z1 ' each to the ZS ', wherein variable Z1 ' to ZS ' corresponding to these assessment expression formulas candidate.Note, more specifically,, when operating position surpasses upper limit quantity Kmax for the quantity of " using ", do not obtain linear combination candidate and AIC for variable Zs '.

In step S201, it is the variable Zs ' of minimum value that selected cell 116 is selected the AIC that obtains to the ZS ' at variables corresponding Z1 '.Here, in the assessment expression formula candidate and linear combination coefficient candidate that obtain to ZS ' about variable Z1 ', assessment expression formula and linear combination coefficient that the assessment expression formula candidate Jv ' of selected variable Zs ' and linear combination coefficient candidate Bsq are best suited for.

In step S202, whether the AIC that selected cell 116 is determined selected variable Zs ' is less than the minimum value of the AIC in the executed processing so far.For example, the AIC in the executed so far processing of selected cell 116 maintenances is variable Zs ', the AIC about this variable Zs ', assessment expression formula candidate Jv ' and the linear combination coefficient candidate Bsq of minimum value.Subsequently, selected cell 116 is with the minimum value of the AIC of this moment, and the AIC of the variable Zs ' that promptly selects in step S201 and the AIC that is kept compare.

In step S202, determine under the situation of AIC of selected variable Zs ' less than the minimum value of the AIC in the executed processing so far, in step S203, combination coefficient computing unit 114 with selected variable Zs ' as new variable Z.Subsequently, after selected variable Zs ' is used as new variable Z, handles and turn back to step S192, repeat above-mentioned processing herein.

Particularly, the AIC of selected variable Zs ' less than before the situation of AIC under, compared with the linear combination coefficient candidate with former assessment expression formula candidate, assessment expression formula candidate Jv ' and the linear combination coefficient candidate Bsq of the variable Zs ' that obtain this moment are assessment expression formula and the linear combination coefficients that is more suitable for.Therefore, there is a kind of possibility, the i.e. AIC of the new variable Zs ' that operating position obtained by the revising Partial Variable Zs ' AIC of the variable Zs ' that selects much smaller than this moment that becomes, and obtain to be more suitable for assessment expression formula candidate Jv ' and linear combination coefficient candidate Bsq as assessment expression formula and linear combination coefficient.

Therefore, assessment expression formula selected cell 94 repeats above-mentioned processing and no longer improves along with the variable Zs ' of the new variable Z of selected conduct this moment up to AIC, i.e. the minimum value of AIC before the AIC that determines selected variable Zs ' is equal to or greater than.

On the other hand, under the situation of the minimum value of the AIC before the AIC of definite selected variable Zs ' is equal to or greater than in step S202, for selected variable Zs ' as new variable Z, even obtained assessment expression formula candidate and linear combination coefficient candidate, the possibility that AIC is modified is also less, therefore handles proceeding to step S204.

In step S204, selected cell 116 output wherein in executed processing so far AIC become the assessment expression formula candidate Jv ' of variable Zs ' of minimum value and linear combination coefficient candidate Bsq as follow-on assessment expression formula and linear combination coefficient.

Particularly, selected cell 116 will for AIC be minimum value V of variable Zs ' assessment expression formula candidate J1 ' to JV ' as follow-on assessment expression formula, and with obtained for the linear combination coefficient candidate Bsq of variable Zs ' as follow-on linear combination coefficient.At this moment, (wherein the user of 1≤n≤N) belong to organizes Gq (the linear combination coefficient candidate Bsq of 1≤q≤Q) wherein, promptly linear combination coefficient bsq0 to bsqV is used as the follow-on linear combination coefficient of user Un user Un.

The follow-on assessment expression formula of Huo Deing is provided to candidate's generation unit 91 and assessment expression formula holding unit 24 from assessment expression formula selected cell 94 like this, and each user's follow-on linear combination coefficient is provided to assessment expression formula holding unit 24 from assessment expression formula selected cell 94.

After having exported follow-on assessment expression formula and linear combination coefficient, the assessment expression formula selects processing to finish, and handles the step S85 that proceeds among Figure 19.

Like this, unit 23 generates assessment expression formula candidate, and supposes that some assessment expression formula candidates are used as follow-on assessment expression formula.Subsequently, unit 23 is carried out these assessment expression formula candidates and corresponding to the evaluation between this assessment expression formula candidate's the linear combination coefficient candidate, and select to be used to assess the user for the assessment expression formula candidate who is more suitable for of the evaluation of content and linear combination coefficient candidate as follow-on assessment expression formula and linear combination coefficient.

Like this, the assessment expression formula candidate and the linear combination coefficient candidate that are more suitable for selected in generation by repeatedly carrying out assessment expression formula candidate and the evaluation of carrying out assessment expression formula candidate and linear combination coefficient candidate, can obtain to assess expression formula and linear combination coefficient thus, this can assess the evaluation of user for content more accurately each the execution when assessment expression formula is selected to handle.

Particularly,, can easily set up the collaborative filtering that is more suitable for, promptly assess expression formula, and can easily improve the assessment accuracy of evaluation in order to assess the evaluation of user for content.Carry out assessment expression formula candidate's combination and linear combination coefficient candidate's evaluation, assessment expression formula candidate that selection is more suitable for and linear combination coefficient candidate can further improve the assessment accuracy of user for the evaluation of content thus as follow-on assessment expression formula and linear combination coefficient.Therefore, can set up a kind of proposed algorithm, wherein with higher accuracy, easier and promptly assess evaluation for the user.

Simultaneously, according to unit 23, when setting up proposed algorithm, promptly when generating assessment expression formula and linear combination coefficient, user (keeper) needn't executable operations.Especially, under the situation that the part of proposed algorithm need be revised, for example, even reappraised under the situation of content the user and to be added to the evaluation objective situation in new content inferior, handle according to study, do not having to obtain suitable assessment expression formula and linear combination coefficient under the situation of manpower by unit 23.That is to say that the keeper does not need to rebulid proposed algorithm.

Notice that in the explanation of carrying out so far, the quantity K of assessment expression formula is equal to or less than upper limit quantity Kmax.Therefore, be under 1 the situation at the quantity K of assessment expression formula, generate the linear combination coefficient bn1 that will multiply each other for interim user's prediction matrix Tn1, multiply by the interim prediction matrix of linear combination coefficient, i.e. the linear combination coefficient bn0 of prediction and evaluation value addition.Simultaneously, be that linear combination coefficient bn0 can be set to 0 consistently under 1 the situation at the quantity K of assessment expression formula.

In addition, be under 1 the situation at the quantity K of assessment expression formula, can be by not using linear combination coefficient but use assessment expression formula F1 separately and obtain the prediction and evaluation matrix.In this case, estimate the evaluation that matrix and assessment result Hs carry out assessment expression formula candidate with the user, and in a plurality of assessment expression formula candidates, the prediction and evaluation value that obtains with assessment expression formula candidate and actual before error between the evaluation of estimate of input more little, then its evaluation is high more.

Simultaneously, above-mentioned a series of processing not only can be carried out and can also carry out by software by hardware.Carrying out by software under the situation of a series of processing, the program that constitutes this software is installed to the computer that embeds the specialized hardware from program recorded medium, or is installed to the general purpose personal computer etc. that the polytype program that can pass through to be installed is carried out polytype function.

Figure 28 shows the block diagram of the ios dhcp sample configuration IOS DHCP of computer, and above-mentioned a series of processing are carried out by program in this computer.For computer, CPU (CPU) 201, ROM (read-only memory) 202 and RAM (random access memory) 203 are interconnected by bus 204.

Input/output interface 205 further is connected to bus 204.The output unit 207 that input/output interface 205 is connected with the input unit 206 that is made of keyboard, mouse, microphone etc., be made of display, loud speaker etc., the record cell 208 that constitutes by hard disk, nonvolatile memory etc., the communication unit 209 that constitutes by network interface etc. and be used to drive for example driver 210 of removable mediums 211 such as disk, CD, magneto-optical disk, semiconductor memory.

For the computer of such configuration, carry out above-mentioned a series of processing, for example, be loaded among the RAM 203 by input/output interface 205 and bus 204 the program in the record cell 208 of will being stored in and carry out this program by CPU 201.

The program of being carried out by computer (CPU 201) can provide by being recorded in the removable medium 211, wherein removable medium 211 is the encapsulation mediums that are made of disk (comprising floppy disk), CD (CD-ROM (compact disc read-only memory), DVD (Digital video disc) etc.), magneto optical disk, semiconductor memory etc., or provides by the wired or wireless transmission medium of for example local area network (LAN), internet or digital satellite broadcasting.

Subsequently, movably medium 211 is installed on the driver 210, and program can be installed in the record cell 208 by input/output interface 205 thus.Simultaneously, program can receive at communication unit 209 places by wired or wireless transmission medium, and is installed in the record cell 208.Replacedly, program can be installed among the ROM 202 or in the record cell 208 in advance.

Notice that the program of being carried out by computer can be wherein to carry out the program of handling by the order that illustrates in this specification with sequential, perhaps can be concurrently or the suitable moment as when execution is called, carry out the program of handling.Be noted that simultaneously embodiments of the invention are not limited to the foregoing description, and under the situation that does not deviate from the spirit and scope of the invention, can carry out multiple modification.

The application comprises about disclosed theme among the Japanese priority patent application JP 2008-111119 that is submitted to Japan Patent office on April 22nd, 2008, and the full content of this patent application is incorporated herein by reference.

Those skilled in the art should understand that according to designing requirement and other factors and can in the scope of claims or its equivalent, carry out multiple modification, combination, sub-portfolio and change.

Claims

1. assessment apparatus comprises:

The prediction and calculation device, be configured to assess not the described evaluation of estimate of carrying out the described evaluation objective estimated by described user based on the evaluation matrix that constitutes for each the evaluation of estimate of evaluation in a plurality of evaluation objectives by expression each among a plurality of users and assessment expression formula, and obtain the prediction and evaluation value as the described evaluation of estimate of assessment, described assessment expression formula is used for assessing described evaluation of estimate by the calculating of using described evaluation matrix; And

Linear combination device, be configured to by using linear combination coefficient, obtain thus for the final assessment result of evaluation of not carrying out the described evaluation objective of described evaluation by described user to carrying out linear combination by using a plurality of described prediction and evaluation that a plurality of described assessment expression formulas are obtained.

2. assessment apparatus according to claim 1, wherein said assessment expression formula is made of a plurality of operators, and described operator comprises carries out the operator that collaborative filtering calculates.

3. assessment apparatus according to claim 2 also comprises:

Assessment expression formula candidate generating apparatus, be configured to described a plurality of described assessment expression formulas as assessment expression formula candidate, described assessment expression formula candidate is a more recent application in the candidate of a plurality of described assessment expression formula of calculating described final assessment result, and this assessment expression formula candidate generating apparatus is configured to generate new arbitrarily assessment expression formula and assesses the new assessment expression formula of the part acquisition of expression formulas and be used as described assessment expression formula candidate by revising in described a plurality of described assessment expression formula some;

The assessment result generating apparatus, be configured to, at each described assessment expression formula candidate, based on described assessment expression formula candidate and described evaluation matrix, calculate each described user in the described evaluation matrix for each the described prediction and evaluation value in described a plurality of described evaluation objectives, generate the assessment result that constitutes by the described prediction and evaluation value that obtains by calculating;

The linear combination coefficient calculation element, be configured to some the assessment results in a plurality of described assessment results as use, and obtain described linear combination coefficient by assessment result and the described evaluation matrix of using described use the assessment expression formula candidate who use to use under as the situation of described assessment expression formula, the assessment expression formula candidate of described use is the described assessment expression formula candidate who is used to generate the assessment result of described use;

Evaluating apparatus is configured to calculate as the amount of information benchmark for the evaluation of the assessment expression formula candidate of described use and described linear combination coefficient; And

Choice device is configured to select to have in the assessment expression formula candidate of described use and the described linear combination coefficient assessment expression formula candidate of described use of high praise as newly being applied to described a plurality of described assessment expression formula and the described linear combination coefficient that described final assessment result is calculated according to described amount of information benchmark.

4. assessment apparatus according to claim 3, wherein, described linear combination coefficient calculation element, in the described a plurality of described users that belong to a group in a plurality of groups, use to belong to and obtain each described linear combination coefficient of described group with the described user's of described group of identical group described evaluation of estimate and described prediction and evaluation value.

5. appraisal procedure that is used for assessment apparatus, this method comprises:

The prediction and calculation means, be configured to assess not the described evaluation of estimate of carrying out the described evaluation objective estimated by described user based on the evaluation matrix that constitutes for each the evaluation of estimate of evaluation in a plurality of evaluation objectives by expression each among a plurality of users and assessment expression formula, and obtain prediction and evaluation value as the described evaluation of estimate of assessment,, described assessment expression formula is used for assessing described evaluation of estimate by the calculating of using described evaluation matrix;

The linear combination means, be configured to by using linear combination coefficient, obtain thus for the final assessment result of evaluation of not carrying out the described evaluation objective of described evaluation by described user to carrying out linear combination by using a plurality of described prediction and evaluation value that a plurality of described assessment expression formulas are obtained;

Described prediction and calculation means are based on described assessment expression formula and described evaluation matrix and obtain described prediction and evaluation value for described a plurality of described assessment expression formulas; And

Described linear combination means make each the described prediction and evaluation value that obtains at each described assessment expression formula carry out linear combination by using described linear combination coefficient, to obtain the described final assessment result of described user for the evaluation of described evaluation objective.

6. one kind makes computer carry out the program of processing, and this processing may further comprise the steps:

Assess not the described evaluation of estimate of carrying out the described evaluation objective estimated by described user based on the evaluation matrix that constitutes for each the evaluation of estimate of evaluation in a plurality of evaluation objectives by expression each among a plurality of users and assessment expression formula, and obtain the prediction and evaluation value as the described evaluation of estimate of assessment, described assessment expression formula is used for assessing described evaluation of estimate by the calculating of using described evaluation matrix; And

By using linear combination coefficient, obtain thus for the final assessment result of evaluation of not carrying out the described evaluation objective of described evaluation by described user to carrying out linear combination by using a plurality of described prediction and evaluation value that a plurality of described assessment expression formulas obtain.

7. assessment apparatus, it comprises:

Prediction and calculation unit, be configured to assess not the described evaluation of estimate of carrying out the described evaluation objective estimated by described user based on the evaluation matrix that constitutes for each the evaluation of estimate of evaluation in a plurality of evaluation objectives by expression each among a plurality of users and assessment expression formula, and obtain the prediction and evaluation value as the described evaluation of estimate of assessment, described assessment expression formula is used for assessing described evaluation of estimate by the calculating of using described evaluation matrix; And

The linear combination unit, be configured to by using linear combination coefficient, obtain thus for the final assessment result of evaluation of not carrying out the described evaluation objective of described evaluation by described user to carrying out linear combination by using a plurality of described prediction and evaluation that a plurality of described assessment expression formulas obtain.

8. appraisal procedure that is used for assessment apparatus comprises:

The linear combination unit, be configured to by using linear combination coefficient, obtain thus for the final assessment result of evaluation of not carrying out the described evaluation objective of described evaluation by described user to carrying out linear combination by using a plurality of described prediction and evaluation value that a plurality of described assessment expression formulas obtain;

Described prediction and calculation unit is based on described assessment expression formula and described evaluation matrix and obtain described prediction and evaluation value for described a plurality of described assessment expression formulas; And

Described linear combination unit makes each the described prediction and evaluation value that obtains at each described assessment expression formula carry out linear combination by using described linear combination coefficient, to obtain the described final assessment result of described user for the evaluation of described evaluation objective.